Robust fast methods to classify variable light curves in large sky surveys are becoming increasingly important. While it is relatively straightforward to identify common periodic stars and particular transient events (supernovae, novae, microlensing), there is no equivalent for non-periodic continuously varying sources (quasars, aperiodic stellar variability). In this paper we present a fast method for modeling and classifying such sources. We demonstrate the method using ~ 86,000 variable sources from the OGLE-II survey of the LMC and ~ 2,700 mid-IR selected quasar candidates from the OGLE-III survey of the LMC and SMC. We discuss the location of common variability classes in the parameter space of the model. In particular we show that quasars occupy a distinct region of variability space, providing a simple quantitative approach to the variability selection of quasars.